Journal of Number Theory 133 (2013) 3453–3469

Contents lists available at SciVerse ScienceDirect

Journal of Number Theory

www.elsevier.com/locate/jnt

Ramanujan type 1/π approximation formulas

N.D. Bagis a,∗ , M.L. Glasser b
a
Stenimahou 5 Edessa Pellas 58200, Greece
b
Department of Theoretical Physics, Faculty of Science, University of Valladolid, Valladolid 40007, Spain

a r t i c l e i n f o a b s t r a c t

Article history: In this article we use analytic and numerical methods to evaluate,
Received 2 November 2012 in closed form, the parameters in various Ramanujan type 1/π
Revised 9 April 2013 formulas.
Accepted 9 April 2013
© 2013 Elsevier Inc. All rights reserved.
Available online 13 June 2013
Communicated by David Goss

Keywords:
π -formulas
Ramanujan
Elliptic functions
Singular modulus
Alternative modular bases
Numerical methods

1. Introduction

During his short life the Indian mathematician Ramanujan (1887–1920) bequeathed to us a great
deal of useful and significant work. Since his death, and especially in the last few decades, many great
mathematicians around the world have been studying his legacy. Most of his results have finally been
proven and set in order, except for a few gaps. It is relevant here to quote B. Berndt from his book [4].

. . . The primary disadvantage of this method (meaning our proofs), as well as most elementary
algebraic approaches, is that the modular equation must be known in advance. Thus the proofs are
perhaps more aptly called verifications.

In spite of all this effort, we are just beginning to understand and organize certain aspects of his
work.

* Corresponding author.
E-mail addresses: [email protected] (N.D. Bagis), [email protected] (M.L. Glasser).

0022-314X/$ – see front matter © 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jnt.2013.04.010
3454 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469

Modular equations are algebraic relations between certain algebraic valued functions. The canon-
ical form of these equations are polynomials which are, in general, not solvable except for those of
lower degrees. Examples occur with respect to the elliptic singular modulus kr , the Rogers–Ramanujan
continued fraction R (q), ratios of theta functions and other special functions. The transformation
(polynomial relation) of an n-th valued modular function to p 2 n is called the modular equation of
degree p.
In his famous paper ‘Modular equations and approximations to π ’ Ramanujan developed a theory
for the construction of series converging to 1/π . These series are rooted in modular equations. More
precisely he developed relations of the form

∞

1 (s)n ( 1 )n (1 − s)n (x + n)
= 2
(a + bn) zn , where (x)n = (1)
π (n!)3 (x)
n =0

where (x) is the well-known Euler Gamma function and s takes the values 2, 3, 4 and 6. Other
values of s do not give us modular bases or they lead to other, known, cases. The parameters a, b and
z are algebraic numbers and can be found by using modular equations.
Interest in formulas like (1) is high at this time for both theoretical and practical purposes; e.g.
they have been used to evaluate the first billions of digits of π (see [22]). In the past mathemati-
cians independently developed related theories which are useful for assessing formulas such as (1):
Eisenstein series, Dirichlet L functions, hypergeometric functions and quadratic class fields.
In this article we shall make use of the elliptic singular modulus kr and elliptic alpha function a(r ),
to motivate and investigate the general form of these formulas. For further information see [2–5,7,9,
11,14,16,18,21].
Next we continue with some definitions and notation. We will basically need only a few simple
and widely known results.
We define the elliptic integrals of the first and second kind respectively (see [4,9,8,13,14]):

π /2 π /2
dt
K (x) =  and E (x) = 1 − x2 sin2 (t ) dt . (2)
2
0 1 − x2 sin (t ) 0

For numerical evaluations and verifications, in the notation of Mathematica, we have

   
K (x) = EllipticK x2 and E (x) = EllipticE x2 . (3)

Also we have (see [7,9]):

dK (x) E (x) K (x)

K (1) (x) = = − . (4)
dx x(1 − x2 ) x

The elliptic singular modulus kr is defined by means of Weber’s functions (see [9,14]):

∞
 ∞

 
G r = 2−1/4 q−1/24 1 + q2n−1 , gr = 2−1/4 q−1/24 1 − q2n−1 (5)
n =1 n =1

as

 −1/12   −2 −1/12
G r = 2kr kr , gr = 2kr kr (6)
 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469 3455

and is a solution of the equation

K (kr ) √
= r. (7)
K (kr )

The function kr = 1 − kr2 is called the complementary singular modulus. When r is rational then kr
is an algebraic number. In Mathematica the singular modulus kr is accessed by the command

 √ 
w = k = kr = k[r ] = InverseEllipticNomeQ e −π r 1/2
. (8)

Also, we will need the following relation for the elliptic alpha function (see [7]):
 
π √ E (kr )
a(r ) = − r −1 . (9)
4K (kr )2 K (kr )

The hypergeometric functions (see [1,3]) are defined by

∞
 (a1 )s (a2 )s . . . (am )s z s
m F n (a1 , a2 , . . . , am ; b 1 , b 2 , . . . , bn ; z ) := (10)
(b1 )s (b2 )s . . . (bn )s s!
s =0

and (a)0 := 1, (a)n := a(a + 1)(a + 2) . . . (a + n − 1), for each positive integer n. For m = n + 1 it
converges for | z| < 1.

2. The derivation of some 1/π and 1/π 2 formulas

One has (see [7, p. 178]):

 √
  4K 2 ( 1
(1 − 1 − z))
1 1 1 2
φ1 ( z ) = 3 F 2 , , ; 1, 1; z = , (11)
2 2 2 π2
which is a special case of Clausen’s product for hypergeometric functions given in Slater [23, p. 75].
Consider the following equation satisfied by the function φ1 ( z):

∞
 ( 12 )n3 n g g
z (a1n + b1 ) = ⇔ b1 φ1 ( z) + a1 zφ1 ( z) = .
(n!)3 π π
n =0

 √ √
Setting w = 1
2
(1 − 1 − z ), 1 − 2w 2 = 1 − z = kr we easily get, using (3),

∞
 ( 12 )n3 n  n 4K ( w )(a1 E ( w ) + (b1 − a1 (1 − w ) − 2b1 w ) K ( w ))
4 w − w 2 (a1 n + b1 ) = . (12)
(n!) 3 π 2 (1 − 2w )
n =0

Using now (9) and setting a1 = 1, b1 = ( √ − kr2 )(1 − 2kr2 )−1 we get the following theorem
a(r )
r

Theorem 2.1.

∞  √ 
 ( 12 )n3 n   2n a(r ) − rkr2 1
4 k r k r n + √ = √ . (13)
(n!)3 r (1 − 2kr )
2
π r (1 − 2kr2 )
n =0
3456 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469

Examples.
√ √
1) For r = 2 we have k2 = 2 − 1 and a(2) = 2 − 1, hence we have from relation (13)

∞   √
 ( 12 )n3 √ 2 1 8+5 2
n
(40 2 − 56) n + − √ = .
(n!)3 7 7 2 14π
n =0

2) For r = 13 we have

 1
4k213 1 − k213 = √
649 + 180 13

and


1 √ √
a(13) = ( 13 − −258 + 74 13 ).
2

Hence (13) becomes

 n  
∞ √
( 12 )n3 1 1 1 1 5
√ n+ (39 − 7 13 ) = + √ .
(n!) 649 + 180 13
3 156 π 26 36 13
n =0

An example of evaluation of (12) is obtained by setting r = 2.

∞ √ √
( 12 )n3 4a 5a Γ 2 ( 98 )
(40 2 − 56)n (an + b) = + √ + 4(−4a + 2a + 14b)
(n!) 3 7π 7 2π 7π Γ 2 ( 58 )
n =0

a and b are arbitrary numbers.

Continuing as in the above theorem we have

Theorem 2.2. Set


n
2 j 2 2n−2 j 2
B n := 16−n
(2 )
j n− j
;
j =0

then

∞
 √ √ 1
rkr 2n + a(r ) −
(2 )
B n (kr )2n rkr2 = . (14)
n =0
π

Proof. We use the definition

 2 √
1 1 4K ( z )2
φ2 ( z ) = 2 F 1 , ; 1; z = . (15)
2 2 π2
 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469 3457

Then if

n 
     2 
n
1 n 1 1 2 j 2 2n−2 j 2
(2 )
B n := j = 16−n n− j
(n!)2 2 j 2 n− j
j
j =0 j =0

∞
 n      2
zn  n 1 1
φ2 ( z ) = · · · = j , (16)
(n!)2 2 j 2 n− j
n =1 j =0

and

∞
 
c φ2 ( z) + bzφ2 ( z) + az2 φ2 ( z) =
(2 )
B n zn an2 + (b − a)n + c .
n =0

Hence we get using (4)

∞
 (2 )  2(−3a + 2b − 2c + 2(2a − 2b + 2c )kr2 ) K (kr )2
B n kr2n an2 + (b − a)n + c =
n =0
π 2 (1 − kr2 )
2(aE (kr )2 + (2b − 2bkr2 − 4a + 6akr2 ) E (kr ) K (kr ))
+ .
π 2 (1 − kr2 )2

For a = 0, b = 1, c = (−kr2 + a(r )r −1/2 )kr −2 the proof follows by using (9). 2

Theorem 2.3. Set

1 2 j 3 2n−2 j 3
n
(3 )
B n := j n− j
(17)
43n
j =0

then we have the 1/π 2 series

∞
 (3 )  2n   3
Bn 2kr kr n2 + b(r ) − 1 n + c (r ) = (18)
n =0
(1 − 2kr2 )2 r π 2

where
√ √ √
3a(r ) + r − 6a(r )kr2 − 9 rkr2 + 12 rkr4
b(r ) = √
r (1 − 2kr2 )2

and
√
3a(r )2 − 6a(r ) rkr2 − rkr2 + 4rkr4
c (r ) = .
r (1 − 2kr2 )2

Proof. Set

 2  4K 2 ( 1 (1 − √1 − z ) ) 2
1 1 1 2
φ3 ( z ) = 3 F 2 , , ; 1, 1; z = ,
2 2 2 π2
3458 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469

then

∞
 
c φ3 ( z) + bzφ3 ( z) + az2 φ3 ( z) =
(3 )
B n zn an2 + (b − a)n + c .
n =0

The left part of the above equation is a function of E (x), K (x), and can evaluated using Theorems 2.1
and 2.2 with parameters a = 1, b = b(r ) − 1, c = c (r ). 2

Note. Theorem 2.3 is very interesting, because it is a general formula for 1/π 2 . Examples of such
formulas are rare in the literature and they not presented as generally as here.

Examples.

1) For r = 25 we have

1
k25 = 
,
√ √
103 682 + 46 368 5 + 24 37 325 880 + 16 692 641 5
√
√
5(841 + 376 5 − 2 269 230 + 120 414 5 )
a(25) = √ .
1682 + 752 5

Hence

∞
    
(3 )
√ n 2 521 5 521
B n (51 841 − 23 184 5) n + 1− √ n+ − √
288 5 12 576 5
n =0

1
= √ . (19)
1200(161 5 − 360)π 2

2) For r = 163 in (18) we get a formula for 1/π 2 derived using Mathematica, by implementing the
method presented in Section 5 below.

Note. The reason of selecting r = 163 is because the class number h(−163) is 1 (see relation (51) be-
low) and for small h(−r ) the algebraic routines with Mathematica give solvable minimal polynomials
of the parameters.

191 211 325 848 427 1 010 784 962 625 383 717 350 772 720 · 22/3
b(163) = − √
151 931 373 056 001 151 931 373 056 001( B 1 − 489B 2 )1/3
√
4 · 21/3 ( B 1 − 489B 2 )1/3
− ,
151 931 373 056 001
B 1 = 5 680 848 001 702 137 216 093 843 898 647 314 524 189,
B 2 = 76 896 989 960 589 381 643 149 203 281 167 − 5 839 006 481 108 705 728

+ 9 529 627 071 955 041 072 · b(163) − 4 530 513 053 635 162 884 · b(163)2
+ 668 649 972 819 460 401 · b(163)3 = 0,
 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469 3459

√
14 178 679 829 869 760 4( C 1 − 489C 2 )1/3
c (163) = −
24 764 813 808 128 163 24 764 813 808 128 163
6 241 484 569 597 616 793 758 909 818 952 · 22/3
− √ ,
24 764 813 808 128 163(C 3 − 489C 4 )1/3
C 1 = 5 512 985 602 111 283 751 597 893 407 219 881 834 715 037 026,

C 2 = 101 526 256 966 667 546 381 077 303 112 958 296 550,

C 3 = 2 756 492 801 055 641 875 798 946 703 609 940 917 357 518 513,
C 4 = 50 763 128 483 333 773 190 538 651 556 479 148 275 − 24 380 823 840 878 077 184

+ 13 131 020 889 593 608 594 752 · c (163) − 30 513 780 896 384 581 928 640 · c (163)2
+ 17 765 361 127 840 243 394 169 · c (163)3 = 0,
∞
 (3 )  2n   A
4n B n k163k163 n2 + b(163) − 1 n + c (163) = ,
n =0
π2
√ √
4(12 660 947 754 667 + 26 680( A 1 − 489 A 2 )1/3 + 26 680( A 1 + 489 A 2 )1/3 )
A= ,
8 254 937 936 042 721
A 1 = 106 866 398 697 613 339 845 357 037,

A 2 = 3 136 555 671 686 449 089, (20)

The value of (k163 k163 )2 is evaluated with Mathematica.

 2 1 266 933 400 10 005 √

y 163 = k163k163 = − √ + (−1 + 55 7403 489 )1/3
16 (−1 + 557 403 489 )1/3 2

the minimal polynomial of y 163 is

−1 + 16 408 588 290 048 048 · y 163 − 768 · y 2163 + 4096 · y 3163 = 0.

Formula (20) gives about 17 digits per term. For r = 253 we have found in the same way another
such formula which gives 21 digits per term, but for economy of space we omit it. (The routine and
the formula have been verified with Mathematica.)

3. An unusual 1/π series

In this section we examine the same 1/π formula as in [20]. The results are compared with two
parametric formulas in the cubic and the fifth modular bases (see relations (36) and (39)), presented
for the first time (see [15]).
The j invariant is given by (see [13]):

 16  8 3
η ( z /2 ) η( z)
j ( z) = + 16 , (21)
η( z) η ( z /2 )
√
where z = −r, r-positive real and
3460 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469

∞
 
η(z) = eπ iz/12 1 − e 2π inz
n =1

is the Dedekind eta function.

Also

1/12
η( z) kr
= . (22)
η(z/2) 21/6kr 1/6
√
From [19, Section 7, Theorem 7.4] and from [10, formula (5.8)], when q = e 2π iz , z = −r, r positive
real, the modular j-invariant is given by

Q 3 (q)
j ( z) = 1728 3 (q ) −
, (23)
Q R 2 (q)

where

∞
 ∞

nqn n3 qn
P (q) = 1 − 24 , Q (q) = 1 + 240
1 − qn 1 − qn
n =1 n =1

and

∞
 n5 qn
R (q) = 1 − 504 .
1 − qn
n =1

The function tr is given by

 
Qr 6
tr = Pr − √ , (24)
Rr π r
where

 √  √  √
P r = P −e −π r
, Q r = Q −e −π r
and R r = R −e −π r
.

i) Using Theorems 3 and 4 of [20], the relation (23) can transformed to

432
j ( z) = . (25)
βr (1 − βr )

Also note that we have

√ 256(1 − kr2 + kr4 )3 432

j ( −r ) = j r = = . (26)
(kr kr )4 βr (1 − βr )

Hence with our method in [20] we can simplify the results of [19] and [10]. The function βr is defined
as the root of the equation:

1 5
2 F 1 ( 6 , 6 ; 1; 1 − w ) √
1 5
= r. (27)
2 F 1 ( 6 , 6 ; 1; w )
 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469 3461

ii) Set now mr := kr2 and let a(r ), E (x) be the elliptic alpha function and the complete elliptic
integral of the second kind respectively then:

 
1 6
tr = P (q) − √
(1 − 2βr /4 )ur2/4 rπ
 
1 E (mr /4 ) 3π
= 3 − 2 + mr /4 − √ F r2/4
(1 − 2βr /4 )ur2/4 K (mr /4 ) 4 r /4K (mr /4 )2

with F r = π2 K (kr ). Equivalently,

1 + mr /4 − √6 a( 4r )
r
tr =  . (28)
1 − mr /4 + mr2/4 (1 − 2βr /4 )

From the above evaluations and the 1/π series in [10] we get the next reformulation:

Theorem 3.1. If we define

J r := 1728 j r−1 = 4βr (1 − βr ), (29)

1/3
2 j r σ (r )G r8
T r := √ √ (30)
r j r − 1728

then

∞ 1

3 ( )n ( 5 )n ( 1 )n
6 6 2
√√ = ( J r )n (6n + 1 − T r ). (31)
π r 1 − Jr (n!)3
n =0

3.1. Similarities of (31) and a fifth order base formula

From the identity

 √ 2  
1 5 1− 1−z 1 5 1
2 F1 , ; 1; = 3 F2 , , ; 1, 1; z , (32)
6 6 2 6 6 2

and the following relations found in [7, Chapter 5]:

   
π 1 1 π 1 1
K s (x) = 2 F1 − s, + s; 1; x2 and E s (x) = 2 F1 − − s, + s; 1; x2 , (33)
2 2 2 2 2 2
 k(1 − k2 ) dK s (t )
E s = 1 − k2 K s + K̇ s , K̇ s (t ) = , (34)
1 + 2s dt
 
π cos(π s) √ E s (xr )
a s (xr ) := − r −1 , (35)
4K s (xr )2 1 + 2s K s (xr )

with s = 1/3 one can get, proceeding as in Theorem 2.1, the following Ramanujan-type 1/π formula
for the 5-th modular base (see [6,12,15,17,20]):
3462 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469

Theorem 3.2.

∞ 1
  βr − a5√(βrr ) 
( )n ( 5 )n ( 1 )n 
6 6 2
n 3
4βr (1 − βr ) 3n − 5 = √ . (36)
(n!)3 1 − 2βr 2π r (1 − 2βr )
n =0

The βr are solutions of Eq. (27) and the function α5 (βr ) of relation (35) is algebraic for r ∈ Q∗+ .

Also comparing (36) with (31) leads to the following theorem.

Theorem 3.3.

√  1 + kr2 − 3 √ 
a(r )

r r
a5 ( βr ) = 1 + 8βr −
. (37)
10 1 − kr2 + kr4

The above formula expresses the base 5 elliptic alpha function in terms of the classical alpha
function a(r ).
Also from the cubic theory as for base 5, we have (see [12]):

   √ 2
1 2 1 1 2 1− 1− w
3 F2 , , ; 1, 1; w = 2 F 1 , ; 1; (38)
3 3 2 3 3 2

and we get

Theorem 3.4.

∞ 1 √
 ( )n ( 2 )n ( 1 )n   n 3
3 3 2 2 2
4α 1 − α
3 3 (n − b) = √ , (39)
(n!)3 2π r (1 − 2α32 )
n =0

4(α32 − a3 (α3 )r −1/2 )

b= . (40)
3(1 − 2α32 )

The function α3 = αr is the third alternative base modulus (see [20]).

4. Further examples and evaluations

1) For r = 3 in relation (31) we have

4
J3 = ,
125
5
T3 =
11

and

√ ∞ 1  n  
5 15  ( 6 )n ( 56 )n ( 12 )n 4 6
= 6n + . (41)
11π (n!)3 125 11
n =0
 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469 3463

2) For r = 4 in relation (31) we have

8
J4 = ,
1331
11
T5 = ,
21

11 ∞ 1
  n  
11 3 ( 6 )n ( 56 )n ( 12 )n 8 10
= 6n + . (42)
14π (n!)3 1331 21
n =0

3) For r = 5 in relation (31) we have

1 √
T5 = (139 + 45 5 ),
418
√
27(−1975 + 884 5 )
J5 = .
33 275

Hence

√
21 650 + 5967 5
π
∞  √ 
 ( 16 )n ( 56 )n ( 12 )n −53 325 + 23 868 5 n √
= (836n + 93 − 15 5 ). (43)
(n!) 3 33 275
n =0

4) For r = 8 in relation (31) we have

√  √
k28 = 113 + 80 2 − 4 2(799 + 565 2 ),

√  √ 2
a(8) = 2(10 + 7 2 ) 1 − −2 + 2 2 .

Thus
 √
5
15 2
(84 125 + 81 432 2)
9982π
∞ 1  √   √ 
 ( 6 )n ( 56 )n ( 12 )n 5 643 000 − 3 990 168 2 n 3276 − 1125 2 + 29 946n
= . (44)
(n!)3 1 520 875 4991
n =0

5) For r = 18 we have

√ √
k18 = (−7 + 5 2 )(7 − 4 3 ),
√ √ √
a(18) = −3057 + 2163 2 + 1764 3 − 1248 6,
1 √
α6 =(68 − 27 6 ),
500
√
1 7(49 982 + 4077 6 )
β18 = − √ √ ,
2 10 5(989 + 54 6 )3/2
3464 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469

√
637 326 171 − 260 186 472 6
J 18 = , (45)
453 870 144 125
√
712 075 + 49 230 6
T 18 = . (46)
1 074 514

Hence we get a formula giving 8 digits per term: (Note that the number of digits per term is deter-
mined by the value of J r , approximately.)

√
5 23 124 123 365 − 13 274 820 6
1 074 514π
∞  √ n
 ( 16 )n ( 56 )n ( 12 )n 637 326 171 − 260 186 472 6
=
(n!)3 453 870 144 125
n =0
 √ 
9(40 271 − 5470 6 )
× 6n + . (47)
1 074 514

6) For r = 27

1 1 + 100 · 21/3 − 80 · 22/3
k27 = √ ,
2 2+ 3 − 100 · 21/3 + 80 · 22/3
 
1 √
a(27) = 3 ( 3 + 1) − 21/3
2

(a(27) was obtained from [7, p. 172]).

56 143 116 + 157 058 640 · 21/3 − 160 025 472 · 22/3
J 27 = ,
817 400 375
58 871 825 + 22 512 960 · 21/3 + 13 208 820 · 22/3
T 27 = .
132 566 687

Hence we get the 11 digits per term formula:

935 935
√ √
3(761 257 259 − 157 058 640 2 + 160 025 472 4 )
3 3
π
∞ 1  √ √ 
 ( )n ( 5 )n ( 1 )n 56 143 116 + 157 058 640 3 2 − 160 025 472 3 4 n
6 6 2
=
(n!)3 817 400 375
n =0
 √ √ 
6(12 282 477 − 3 752 160 2 − 2 201 470 4 )
3 3

× 6n + .
132 566 687

7) For r = 58 we have from the Wolfram pages ‘Elliptic Lambda Function’ and ‘Elliptic Singular
Value’:

√ √
k58 = (−1 + 2 )6 (−99 + 13 58 )
 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469 3465

and

1 √ √ √ √
a(58) = (−70 + 99 2 − 13 29 )(5 + 29 )6 (−444 + 99 29 ).
64

Also using the cubic theta identities, (see [20, relations (4), (5), (6) and (33)]) we evaluated α174
numerically to 1500 digits and then β58 to 1500 digits accuracy. We then applied the ‘Recognize’
routine of Mathematica. The result is the minimum polynomial of β58 :

1 − 1 399 837 865 393 267 000x + 79 684 665 286 353 732 299 517 000x2

− 159 369 327 773 031 733 812 500 000x3 + 79 684 663 886 515 866 906 250 000x4 = 0.

Solving this equation with respect to x we obtained the value of β58 in radicals. Thus

√
1 399 837 865 393 267 − 259 943 365 786 104 29
J 58 = , (48)
39 842 331 943 257 933 453 125
√
5(1 684 967 251 + 24 160 612 29 )
T 58 = . (49)
10 376 469 642

The result is the formula

 √
5
5 87
(13 826 969 809 210 107 − 90 211 316 29 )
357 809 298π
∞  √ n
 ( 16 )n ( 56 )n ( 12 )n 1 399 837 865 393 267 − 259 943 365 786 104 29
=
(n!)3 39 842 331 943 257 933 453 125
n =0
 
6 117 973 8 628 790
× − √ + 6n (50)
32 528 118 25 557 807 29

which gives 18 digits per term.

8) For r = 93 (see [7, p. 158]), we have

√ 3
−6 3+1 √ √ √
σ (93) = 6G 93 (15 93 + 13 31 + 201 3 + 217).
2

From [5, Chapter 34] we have

√ √ √
(3 3 + 31 )1/4 (39 + 7 31 )1/6
G 93 =
21/3

also

√ 1 + kr2 σ (r )
a(r ) = r − ,
3 6

G− 24 2 2
93 = 4k93 1 − k93 .
3466 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469

Hence

 2 1
k93 k93 = √ √ √ ,
224 589 314 596 + 129 666 700 800 3 + 40 337 431 680 31 + 23 288 826 960 93
−1
√
J 93 = 119 562 334 956 358 303 022 500 + 21 474 029 280 866 147 440 000 31 + 470 106 000

√
× 129 368 095 019 778 762 513 344 107 725 + 23 235 195 778 655 878 514 048 710 848 31,
√ √ √
10 559 116 299 575 + 1 317 692 448 000 3 + 275 805 228 680 31 − 81 807 235 875 93
T 93 = ,
15 081 520 900 138

where

−1
j 93 = 1728 J 93 .

This results in a very flexible formula that gives about 24 digits per term.

5. A useful Mathematica implementation

The class number h(−d), d ∈ N of equivalent quadratic forms with discriminant d is given by

d −1  
w (d)  −d
h(−d) = − n, (51)
2d n
n =1

n
where w (3) = 6, w (4) = 4 and w (d) = 2 otherwise; ( m ) is the Jacobi symbol. We observe that
h(−163) = 1 (see [13]). For small values of h(−d) we have a greater possibility for evaluating J d
and T d in radicals (see also [21]).
The simplest way to evaluate the parameters J 163 and T 163 is to rely on Mathematica and use the
routine ‘Recognize’.
The general algorithm is:
√
i) Set r = d and k[r ] = InverseEllipticNomeQ[e −π r ]1/2 , then we can evaluate βr and j r from rela-
tions (21) and (25). Hence we get the value of J r as in example 7) of Section 4.
ii) For the evaluation of T r we need the value of a(r ) in relation (9)

 
π √ E
a(r ) = − r −1 . (52)
4K 2 K

This in Mathematica is given from

 
π √ EllipticE[k[r ]2 ]
a(r ) = − r −1 . (53)
4EllipticK[k[r ]2 ]2 EllipticK[k[r ]2 ]

Hence loading the package

 NumberTheory‘Recognize’

and
 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469 3467

 
Recognize N [ J 163 , 1500], 16, x ,
 
Recognize N [ T 163 , 1500], 16, x

we get two equations. After solving them we get, if r ∈ N (here r = 163), the values of the parameters
J r and T r in algebraic form. Thus results in the π formulas.

1) We have that J 163 is root of

−64 + 2 552 810 853 189 232 588 558 727 380 998 000x
− 2 198 253 790 246 041 723 377 943 360 187 500x2
+ 224 451 422 498 574 115 473 590 775 022 822 688 001 953 125x3 = 0

hence
√ √
C 1 − C 2 (− A 1 + 489B 1 )−1/3 + 30 591 288(− A 1 + 489B 1 )1/3
J 163 = 4
10 792 555 251 621 895 860 488 211 571 345 343 375
A 1 = 12 737 965 652 562 547 164 590 026 038 483 234 248 161 827 096 523 072 256 574 968 383,

B 1 = 229 038 073 182 066 825 378 006 485 964 950 394 558 349 727 761 749 294 205 546 402 325 349,

C 1 = 8 808 429 913 332 498 766 352 891,

C 2 = 902 206 261 147 132 595 923 169 636 910 570 558 029 813 352 485 594 880.

From J r = 4βr (1 − βr ), we get the value of βr and hence

√ √
12 948 195 754 365 757 115 + 8( A 2 − B 2 489 )1/3 + 8( A 2 + B 2 489 )1/3
T 163 = 5
83 470 787 671 093 501 833

where

A 2 = 3 802 386 862 487 392 962 897 493 239 274 992 371 253 057 854 289 262,

B 2 = 3 865 464 212 119 923 579 732 688 315 287 754 932 290 919 450.

The above parameters give 32 digits per term.

2) Another evaluation which we consider is taking d = r = 253. This number lead us to non-
trivial-expected evaluation because of good behavior of the J 253 and T 253 , which are produce solvable
equations and is not indicated in the literature [21].

√  √
A 1 − A 2 11 + 31 990 140 A 3 − A 4 11
J 253 = ,
A5
A 1 = 2 804 365 789 259 959 094 417 576 921 792 857 440 357 087 269 234 369,

A 2 = 845 548 099 807 651 569 627 713 349 319 558 464 492 321 957 799 872,
A 3 = 14 334 626 424 019 721 997 733 410 517 481 729 654 402 717 977 139

516 818 782 945 906 676 740 858 207 407 330 990 565,
3468 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469

A 4 = 4 322 052 487 126 125 954 073 317 286 237 053 746 613 433 493 632 282

233 926 553 935 879 770 457 716 659 641 968 088,
A 5 = 1 066 755 353 338 783 886 372 226 117 351 012 749 877 681 799 897 625

and

√ √
1875 B 1 − B 2 11 + 3 847 208 393 012 364 625 + 752 271 279 708 923 520 11
T 253 = ,
6 969 874 104 047 710 086
B 1 = 213 216 899 528 167 866 600 672 118 125,

B 2 = 60 533 150 139 616 794 053 500 831 192.

The above parameters give 41 digits per term.

5.1. Conclusion

We have given a way to construct a very large number of Ramanujan-type 1/π formulas. We have
also presented, perhaps the first, general parametric formula for a class of 1/π 2 series. The formulas
in 3rd and 5th modular bases also appear to be new. It is also demonstrated that in most cases, from
r = 1 to 100 and higher, using Mathematica the construction of such formulas is very simple, as long
as the class number h(−d) remains small and the parameters are solutions of solvable polynomial
equations.

Acknowledgments

The work of M.L. Glasser has been partially supported by the Spanish Ministerio de Educación y
Ciencia (Project MTM 2005-09183) and he thanks the University of Valladolid for continuing hospi-
tality. The authors thank Prof. J. Guillera for discussions.

References

[1] M. Abramowitz, I.A. Stegun, Handbook of Mathematical Functions, Dover Publications, 1972.
[2] B.C. Berndt, Ramanujan’s Notebooks Part I, Springer Verlag, New York, 1985.
[3] B.C. Berndt, Ramanujan’s Notebooks Part II, Springer Verlag, New York, 1989.
[4] B.C. Berndt, Ramanujan’s Notebooks Part III, Springer Verlag, New York, 1991.
[5] B.C. Berndt, Ramanujan’s Notebooks Part V, Springer Verlag, Inc., New York, 1998.
[6] B.C. Berndt, Heng Huat Chan, Ramanujan and the modular j-invariant, Canad. Math. Bull. 42 (4) (1999) 427–440.
[7] J.M. Borwein, P.B. Borwein, Pi and the AGM, John Wiley and Sons Inc., New York, Chichester, Brisbane, Toronto, Singapore,
1987.
[8] I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series and Products, Academic Press, 1980.
[9] E.T. Whittaker, G.N. Watson, A Course on Modern Analysis, Cambridge Univ. Press, 1927.
[10] B.C. Berndt, Heng Huat Chan, Eisenstein series and approximations to π , page stored in the Web.
[11] S. Ramanujan, Modular equations and approximations to π , Quart. J. Math. (Oxford) 45 (1914) 350–372.
[12] B.C. Berndt, S. Bhargava, F.G. Garvan, Ramanujan’s theories of elliptic functions to alternative bases, Trans. Amer. Math.
Soc. 347 (1995) 4163–4244.
[13] D. Broadhurst, Solutions by radicals at singular values k N from new class invariants for N ≡ 3 mod 8, arXiv:0807.2976
[math-phy].
[14] J.V. Armitage, W.F. Eberlein, Elliptic Functions, Cambridge Univ. Press, 2006.
[15] N.D. Baruah, B.C. Berndt, Ramanujan series for 1/π arising from his cubic and quartic theories of elliptic functions, J. Math.
Anal. Appl. 341 (2008) 357–371.
[16] B.C. Berndt, Ramanujan’s theory of theta-functions, in: Maruti Ram Murty (Ed.), Theta Functions: From the Classical to the
Modern, American Mathematical Society, 1993.
[17] J.M. Borwein, P.B. Borwein, A cubic counterpart of Jacobi’s identity and the AGM, Trans. Amer. Math. Soc. 323 (2) (1991)
691–701.
[18] Habib Muzaffar, Kenneth S. Williams, Evaluation of complete elliptic integrals of the first kind at singular moduli, Taiwanese
J. Math. 10 (6) (2006) 1633–1660.
 N.D. Bagis, M.L. Glasser / Journal of Number Theory 133 (2013) 3453–3469 3469

[19] B.C. Berndt, Aa Ja Yee, Ramanujans contributions to Eisenstein series, especially in his lost notebook, page stored in the
Web.
[20] N.D. Bagis, M.L. Glasser, Conjectures on the evaluation of alternative modular bases and formulas approximating 1/π ,
J. Number Theory 132 (2012) 2353.
[21] J.M. Borwein, P.B. Borwein, Class number three Ramanujan type series for 1/π , J. Comput. Appl. Math. 46 (1993) 281–290.
[22] J.M. Borwein, P.B. Borwein, D.H. Bailey, Ramanujan Modular equations, and approximations to Pi or how to compute one
billion digits of Pi, Amer. Math. Monthly 96 (1989) 201–219.
[23] L.J. Slater, Generalized Hypergeometric Functions, Cambridge Univ. Press, London, 1966.

Further reading

[24] I.J. Zucker, The summation of series of hyperbolic functions, SIAM J. Math. Anal. 10 (1979) 192.
[25] S. Chowla, Series for 1/ K and 1/ K 2 , J. Lond. Math. Soc. 3 (1928) 9–12.
[26] N.D. Baruah, B.C. Berndt, H.H. Chan, Ramanujan’s series for 1/π : A survey, Amer. Math. Monthly 116 (2009) 567–587.
[27] T. Apostol, Modular Functions and Dirichlet Series in Number Theory, Springer, 1990.
[28] N.D. Baruah, B.C. Berndt, Eisenstein series and Ramanujan-type series for 1/π , Ramanujan J. 23 (2010) 17–44.
[29] J. Guillera, Series de Ramanujan (Generalizaciones y conjecturas), Thesis, Universidad de Zaragosa, Spain, 2007.
[30] J. Guillera, M. Rogers, Ramanujan series upside-down, arXiv:1206.3981v1 [math.NT], 2012.